The mv643xx_eth mii bus implementation uses wait_event_timeout() to
wait for SMI completion interrupts.
If wait_event_timeout() would return zero, mv643xx_eth would conclude
that the SMI access timed out, but this is not necessarily true --
wait_event_timeout() can also return zero in the case where the SMI
completion interrupt did happen in time but where it took longer than
the requested timeout for the process performing the SMI access to be
scheduled again. This would lead to occasional SMI access timeouts
when the system would be under heavy load.
The fix is to ignore the return value of wait_event_timeout(), and
to re-check the SMI done bit after wait_event_timeout() returns to
determine whether or not the SMI access timed out.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
mv643xx_eth uses ip_hdr() (defined in linux/ip.h), but relied on
another header file to include the needed header file indirectly.
In latest net-next this indirect include chain is gone, so the
driver fails to build. Include linux/ip.h explicitly to fix this.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch introduces mdiobus_alloc() and mdiobus_free(), and
makes all mdio bus drivers use these functions to allocate their
struct mii_bus'es dynamically.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Acked-by: Andy Fleming <afleming@freescale.com>
In preparation of giving mii_bus objects a device tree presence of
their own, rename struct mii_bus's ->dev argument to ->parent, since
having a 'struct device *dev' that points to our parent device
conflicts with introducing a 'struct device dev' representing our own
device.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Acked-by: Andy Fleming <afleming@freescale.com>
This gives a nice increase in the maximum loss-free packet forwarding
rate in routing workloads.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Switch mv643xx_eth from using drivers/net/mii.c to using phylib.
Since the mv643xx_eth hardware does all the link state handling and
PHY polling, the driver will use phylib in the "Doing it all yourself"
mode described in the phylib documentation.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Acked-by: Andy Fleming <afleming@freescale.com>
If we don't poll the hardware statistics counters at least once every
~34 seconds, overflow might occur without us noticing. So, set up a
timer to poll the statistics counters at least once every 30 seconds.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
When the IP header doesn't start 14, 18, 22 or 26 bytes into the packet
(which are the only four cases that the hardware can deal with if asked
to do IP checksumming on transmit), invoke the software checksum helper
instead of letting the packet go out with a corrupt checksum inserted
into the packet in the wrong place.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
We have to explicitly tell the hardware to include the pseudo-header
when doing receive checksumming, otherwise hardware checksumming will
fail for every received packet and we'll end up setting CHECKSUM_NONE
on every received packet.
While we're at it, when skb->ip_summed is set to CHECKSUM_UNNECESSARY
on received packets, skb->csum is supposed to be undefined, and thus
there is no need to set it.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Add support for mv643xx_eth versions that have no transmit bandwidth
control registers at all, such as the ethernet block found in the
Marvell 88F6183 ARM SoC.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Currently, the receive processing reads ->byte_cnt twice (once to
update interface statistics and once to properly size the data area
of the received skb), but since receive descriptors live in uncached
memory, caching this value in a local variable saves one uncached
access, and increases routing performance a tiny little bit more.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Since the size of the receive queue is directly related to the data
cache footprint of the driver (between refilling a receive ring entry
with a fresh skb and receiving a packet in that entry, queue_size - 1
other skbs will have been touched), shrink the default receive queue
size to a saner number of entries, as 400 is definite overkill for
almost all workloads.
While we are at it, trim the default transmit queue size a bit as well.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Get rid of the skb pointer array that we currently use for transmit
reclaim, and replace it with an skb queue, to which skbuffs are appended
when they are passed to the xmit function, and removed from the front
and freed when we do transmit queue reclaim and hit a descriptor with
the 'owned by device' bit clear and 'last descriptor' bit set.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
By moving DMA unmapping during transmit reclaim back under the netif
tx lock, we avoid the situation where we read the DMA address and buffer
length from the descriptor under the lock and then not do anything with
that data after dropping the lock on platforms where the DMA unmapping
routines are all NOPs (which is the case on all ARM platforms that
mv643xx_eth is used on at least).
This saves two uncached reads, which makes a small but measurable
performance difference in routing benchmarks.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Since our ->hard_start_xmit() method is already called under spinlock
protection (the netif tx queue lock), we can simply make that lock
cover the private transmit state (descriptor ring indexes et al.) as
well, which avoids having to use a private lock to protect that state.
Since this was the last user of the driver-private spinlock, it can
be killed off.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Move link status handling, transmit reclaim and TX_END handling from
the interrupt handler to the napi poll handler. This allows switching
->lock over to a non-IRQ-safe lock and removes all explicit interrupt
disabling from the driver.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
As all the infrastructure for multiple transmit queues already exists
in the driver, this patch is entirely trivial.
The individual transmit queues are still serialised by the driver's
per-port private spinlock, but that will disappear (i.e. be replaced
by the per-subqueue ->_xmit_lock) in a subsequent patch.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Delete a couple of unused and uninteresting interrupt source mask bits:
- The receive resource underrun interrupt sources are uninteresting
because if we are in out-of-memory mode, we are already dealing with
the issue, and we don't need the hardware to remind us again that we
are out of memory.
- The LINK and PHY interrupt sources can be coalesced into one define,
since we always use them together.
- The transmit resource underrun interrupt source can be disabled since
we never activate the head descriptor of a paged skb until the
fragments are all activated, so transmit underrun during a packet
should never happen.
- The INT_EXT_TX_0 define is never used.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
There is no need to call netif_{stop,wake}_queue() when the link goes
down/up, as the networking already takes care of this internally.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Currently, there are two different fields in the
mv643xx_eth_platform_data struct that together describe the PHY
address -- one field (phy_addr) has the address of the PHY, but if
that address is zero, a second field (force_phy_addr) needs to be
set to distinguish the actual address zero from a zero due to not
having filled in the PHY address explicitly (which should mean
'use the default PHY address').
If we are a bit smarter about the encoding of the phy_addr field,
we can avoid the need for a second field -- this patch does that.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Which top-level unit's SMI interface to use should be a property of
the top-level unit, not of the individual ports. This patch moves the
->shared_smi pointer from the per-port platform data to the global
platform data.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Simplify receive and transmit queue handling by requiring the set
of queue numbers to be contiguous starting from zero.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Get rid of the mv643xx_eth-internal MV643XX_ETH_CHECKSUM_OFFLOAD_TX
compile-time option. Using transmit checksumming is the sane default,
and anyone wanting to disable it should use ethtool(8) instead of
recompiling their kernels.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
By having the receive out-of-memory handling timer schedule the napi
poll handler and then doing oom processing from the napi poll handler,
all code that touches receive state moves to napi context, letting us
get rid of all explicit locking in the receive paths since the only
mutual exclusion we need anymore at that point is protection against
reentering ourselves, which is provided by napi synchronisation.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Make napi unconditional on the receive side, so that we can get rid
of all the locking and local interrupt disabling in the receive path.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
If the platform code has passed us the IRQ number of the mv643xx_eth
top-level error interrupt, use the error interrupt to wait for SMI
access completion instead of polling the SMI busy bit, since SMI bus
accesses can take up to tens of milliseconds.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Since commit 81600eea98 ("mv643xx_eth:
use auto phy polling for configuring (R)(G)MII interface"),
mv643xx_eth no longer does SMI accesses from interrupt context. The
only other callers that do SMI accesses all do them from process
context, which means we can switch the PHY lock from a spinlock to a
mutex, and get rid of the extra locking in some ethtool methods.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Get rid of the modulo operations that are currently used for
computing successive TX/RX descriptor ring indexes.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Using IRQF_SAMPLE_RANDOM for the mv643xx_eth interrupt handler
significantly increases interrupt processing overhead, so get rid
of it.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
When tearing down a DMA mapping for a receive buffer, we should pass
dma_unmap_single() the exact same address that dma_map_single() gave
us when we originally set up the mapping.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
The mv643xx_eth hardware ignores the lower three bits of the buffer
size field in receive descriptors, causing the reception of full-sized
packets to fail at some MTUs. Fix this by rounding the size of
allocated receive buffers up to a multiple of eight bytes.
While we are at it, add a bit of extra space to each receive buffer so
that we can handle multiple vlan tags on ingress.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
When we are low on memory, the assumption that every descriptor in the
receive ring will have an skbuff associated with it does not hold.
rxq_process() was assuming that if the receive descriptor it is working
on is not owned by the hardware, it can safely be processed and handed
to the networking stack. But a descriptor in the receive ring not being
owned by the hardware can also happen when we are low on memory and did
not manage to refill the receive ring fully.
This patch changes rxq_process()'s bailout condition from "the first
receive descriptor to be processed is owned by the hardware" to "the
first receive descriptor to be processed is owned by the hardware OR
the number of valid receive descriptors in the ring is zero".
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Nicolas Pitre noted that mv643xx_eth_poll was incorrectly using
non-IRQ-safe locks while checking whether to wake up the netdevice's
transmit queue. Convert the locking to *_irq() variants, since we
are running from softirq context where interrupts are enabled.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Commit 12e4ab79cd ("mv643xx_eth: be
more agressive about RX refill") changed the condition for the receive
out-of-memory timer to be scheduled from "the receive ring is empty"
to "the receive ring is not full".
This can lead to a situation where the receive out-of-memory timer is
pending because a previous rxq_refill() didn't manage to refill the
receive ring entirely as a result of being out of memory, and
rxq_refill() is then called again as a side effect of a packet receive
interrupt, and that rxq_refill() call then again does not succeed to
refill the entire receive ring with fresh empty skbuffs because we are
still out of memory, and then tries to call add_timer() on the already
scheduled out-of-memory timer.
This patch fixes this issue by changing the add_timer() call in
rxq_refill() to a mod_timer() call. If the OOM timer was not already
scheduled, this will behave as before, whereas if it was already
scheduled, this patch will push back its firing time a bit, which is
safe because we've (unsuccessfully) attempted to refill the receive
ring just before we do this.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
When a receive interrupt occurs, mv643xx_eth would first process the
receive descriptors and then ACK the receive interrupt, instead of the
other way round.
This would leave a small race window between processing the last
receive descriptor and clearing the receive interrupt status in which
a new packet could come in, which would then 'rot' in the receive
ring until the next receive interrupt would come in.
Fix this by ACKing (clearing) the receive interrupt condition before
processing the receive descriptors.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Although mv643xx_eth has no hardware support for inserting a vlan
tag by twiddling some bits in the TX descriptor, it does support
hardware TX checksumming on packets where the IP header starts {a
limited set of values other than 14} bytes into the packet.
This patch sets mv643xx_eth's ->vlan_features to NETIF_F_SG |
NETIF_F_IP_CSUM, which prevents the stack from checksumming vlan'ed
packets in software, and if vlan tags are present on a transmitted
packet, notifies the hardware of this fact by toggling the right
bits in the TX descriptor.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
When there is a link status change (link or phy status interrupt),
print a message notifying the user of the new link status.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
The mv643xx_eth hardware has a provision for polling the PHY's
MII management registers to obtain the (R)(G)MII interface speed
(10/100/1000) and duplex (half/full) and pause (off/symmetric)
settings to use to talk to the PHY.
The driver currently does not make use of this feature. Instead,
whenever there is a link status change event, it reads the current
link parameters from the PHY, and programs those parameters into
the mv643xx_eth MAC by hand.
This patch switches the mv643xx_eth driver to letting the MAC
auto-determine the (R)(G)MII link parameters by PHY polling, if there
is a PHY present. For PHYless ports (when e.g. the (R)(G)MII
interface is connected to a hardware switch), we keep hardcoding the
MII interface parameters.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
Instead of hardcoding MII register addresses and values, use the
symbolic constants defined in linux/mii.h.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
The mv643xx_eth driver is limiting DMA bursts to 32 bytes, while
using the largest burst size (128 bytes) gives a couple percentage
points performance improvement in throughput tests, and the docs
say that the 128 byte default should not need to be changed, so
use 128 byte bursts instead.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
The recommended sequence for waiting for the transmit path to clear
after disabling all of the transmit queues is to wait for the
TX_FIFO_EMPTY bit in the Port Status register to become set as well
as the TX_IN_PROGRESS bit to clear.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
The maximum receive packet size field in the Port Serial Control
register controls at what size received packets are flagged
overlength in the receive descriptor, but it doesn't prevent
overlength packets from being DMAd to memory and signaled to the
host like other received packets.
mv643xx_eth does not support receiving jumbo frames in 10/100 mode,
but setting the packet threshold to larger than 1522 bytes in 10/100
mode won't cause breakage by itself.
If we really want to enforce maximum packet size on the receiving
end instead of on the sending end where it should be done, we can
always just add a length check to the software receive handler
instead of relying on the hardware to do the comparison for us.
What's more, changing the maximum packet size field requires
temporarily disabling the RX/TX paths. So once the link comes
up in 10/100 Mb/s mode or 1000 Mb/s mode, we'd have to disable it
again just to set the right maximum packet size field (1522 in
10/100 Mb/s mode or 9700 in 1000 Mb/s mode), just so that we can
offload one comparison operation to hardware that we might as well
do in software, assuming that we'd want to do it at all.
Contrary to what the documentation suggests, there is no harm in
just setting a 9700 byte maximum packet size in 10/100 mode, so use
the maximum maximum packet size for all modes.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
The mv643xx_eth driver allows doing transmit reclaim from within the
napi poll routine, but after doing reclaim, it would forget to check
the free transmit descriptor count and wake up the transmit queue if
the reclaim caused enough descriptors for a new packet to become
available. This would cause the netdev watchdog to occasionally kick
in during certain workloads with combined receive and transmit traffic.
Fix this by adding a wakeup check identical to the one in the
interrupt handler to the napi poll routine.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
When the ethernet link goes down while mv643xx_eth is transmitting
data, transmit DMA can stop before all queued transmit descriptors
have been processed. But even the descriptors that _have_ been
processed might not be properly marked as done before the transmit
DMA unit shuts down.
Then when the link comes up again, the hardware transmit pointer
might have advanced while not all previous packet descriptors have
been marked as transmitted, causing software transmit reclaim to
hang waiting for the hardware to finish transmitting a descriptor
that it has already skipped.
This patch forcibly reclaims all packets on the transmit ring on a
link down interrupt, and then resyncs the hardware transmit pointer to
what the software's idea of the first free descriptor is. Also, we
need to prevent re-waking the transmit queue if we get a 'transmit
done' interrupt at the same time as a 'link down' interrupt, which
this patch does as well.
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
The previously merged TX hang erratum workaround ("mv643xx_eth:
work around TX hang hardware issue") assumes that TX_END interrupts
are delivered simultaneously with or after their corresponding TX
interrupts, but this is not always true in practise.
In particular, it appears that TX_END interrupts are issued as soon
as descriptor fetch returns an invalid descriptor, which may happen
before earlier descriptors have been fully transmitted and written
back to memory as being done.
This hardware behavior can lead to a situation where the current
driver code mistakenly assumes that the MAC has given up transmitting
before noticing the packets that it is in fact still currently working
on, causing the driver to re-kick the transmit queue, which will only
cause the MAC to re-fetch the invalid head descriptor, and generate
another TX_END interrupt, et cetera, until the packets in the pipe
finally finish transmitting and have their descriptors written back
to memory, which will then finally break the loop.
Fix this by having the erratum workaround not check the 'number of
unfinished descriptor', but instead, to compare the software's idea
of what the head descriptor pointer should be to the hardware's head
descriptor pointer (which is updated on the same conditions as the
TX_END interupt is generated on, i.e. possibly before all previous
descriptors have been transmitted and written back).
Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
